The Hypernet Relies on infomorphs to maintain its complex superluminal network, and Null Hypothesizers are the most critical of these infomorphs. Their job is to maintain information flow across the Quaternary Membrane.
The technique used to allow information to be transmitted at superluminal speeds while still maintaining USCS Temporal Logic is to assume the knowledge already exists on both sides.
The infomorphs allow their Query Engine to be trained by the information that will be transmitted, and then forget it. Because it is an Ephemeral Indicator system, the message can be largely reconstructed by determining what "surprises" the query engine and what doesn't, even if the engine does not provide an answer.
This cheat works because nearly all of the information already exists in both locations. Nearly all hypernet transmissions are simply mundane updates, and reconstructing them in this fashion doesn't seem to count as introducing any "new" information, thus preserving temporal flow.
Null Hypothesizers often run afoul of quaternary membranes even so, and frequently have to return back and try a different amount of training and forgetting. Of course, they can't really remember what they went through... Because of this, a vast percentage of hypernet storage is taken up with records of who was trained, how, and sent where, when.
One of the largest concerns of people studying the hypernet is how much information already exists on both sides of a superluminal communication.
While it's obvious the hypernet has many cloned databases at each node, it's clear that even without a shared knowledge base the same extremely specific information tends to arise.
It often appears that the universe itself is "echoing" knowledge ahead of time, so as to prepare civilizations for superluminal communication in just this fashion.
However, scientists assure us that's a very silly idea.